# Morning AI TL;DR — 2026-06-02

- **Anthropic filed to go public**, setting up a major public-market test for frontier AI economics and infrastructure burn ([TechCrunch](https://techcrunch.com/2026/06/01/anthropic-files-to-go-public/), [The Verge](https://www.theverge.com/ai-artificial-intelligence/941016/anthropic-has-officially-filed-to-go-public)).
- **Anthropic expanded Claude for Chrome from research preview to beta after publishing prompt-injection defenses**, a useful signal for anyone building browser/CRM agents that touch untrusted pages ([Anthropic](https://www.anthropic.com/news/prompt-injection-defenses), [VentureBeat](https://venturebeat.com/security/anthropic-browser-agent-hijacked-31-percent-before-safeguards-engaged)).
- **JetBrains open-sourced Mellum2, a 12B MoE model with 2.5B active parameters per token**, aimed at efficient code and software-engineering workloads on private infrastructure ([Hugging Face](https://huggingface.co/blog/JetBrains/mellum2-launch), [The New Stack](https://thenewstack.io/jetbrains-mellum2-open-source-coding-model/)).
- **NVIDIA’s local-agent push continued with DGX Spark setup improvements and Vera CPU momentum**, pointing toward more practical on-prem/edge agent deployments rather than cloud-only workflows ([NVIDIA Technical Blog](https://developer.nvidia.com/blog/run-local-ai-agents-with-faster-models-and-multi-node-clustering-on-nvidia-dgx-spark/), [NVIDIA Blog](https://blogs.nvidia.com/blog/vera-cpu-phoronix/)).
- **GitHub Copilot’s usage-based AI Credits shift is now live/discussed**, making cost controls and model-routing strategy more important for teams leaning heavily on agent mode and coding agents ([GitHub Community](https://github.com/orgs/community/discussions/197089), [GitHub Copilot pricing](https://github.com/features/copilot/plans)).
- **Salesforce announced a $2B AI investment in France through 2030**, reinforcing Europe as a strategic Agentforce/Data Cloud growth market and implementation opportunity ([Salesforce](https://www.salesforce.com/news/press-releases/2026/06/01/2-billion-ai-transformation-investment-france/)).
- **Krebs reported attackers abused Meta’s AI support bot to seize Instagram accounts**, a concrete reminder that AI support agents need hard authorization boundaries, not just better prompts ([KrebsOnSecurity](https://krebsonsecurity.com/2026/06/hackers-used-metas-ai-support-bot-to-seize-instagram-accounts/)).

**Worth watching:** The near-term winner for builders may be “secure-by-default agents”: sandboxed execution, explicit auth boundaries, auditable tool use, and predictable usage-based cost controls.